It is early in the year, but the most compelling show for art and tech in 2019 may already be happening. AI artist and Artnome favorite Robbie Barrat has teamed up with renowned French painter Ronan Barrot for a fascinating show that lives somewhere in the margin between collaboration and confrontation.
The L'Avant Galerie Vossen emailed Robbie last April after seeing his AI nude portraits and asked if he would be willing to fly out to Paris to work with Ronan. Robbie agreed and flew out last July to meet with Ronan, and the two have been working together ever since. The show titled “BARRAT/BARROT: Infinite Skulls“ opens Thursday, February 7th, and literally features an “infinite” number of skulls.
Why Skulls?
For the last two decades, it has been artist Ronan Barrot’s tradition to use the remaining paint on his palette to paint a skull each time he stops, interrupts, or finishes a painting. As it was explained to me, the skulls are like a side process of the main painting, it’s like when you clean out your motor after driving for miles and miles. Ronan now estimates that he has painted a few thousand of these, and this massive visual data set of painted skulls was perfect for AI artist Robbie Barrat to use in training his GANs (generative adversarial networks).
GANs are comprised of two neural networks, which are essentially programs designed to think like a human brain. In our case, we can think of these neural networks as being like two people: first, a "generator," whom we will think of as an art forger, and second, as a "discriminator," whom we will think of as the art critic. Now imagine that we gave the art forger a book with 500 skulls painted by Ronan as training material that he could use to create a forgery to fool the critic. If the forger looked at only three or four of Ronan’s paintings, he may not be very good at making a forgery, and the critic would likely figure out the forgery pretty quickly. But after looking at enough of the paintings and trying over and over again, the forger may actually start producing paintings good enough to fool the critic, right? This is precisely what happens with GANs in AI art.
In Robbie’s words:
I trained the network on the skulls. They are all the same shape, the same size, the same orientation, and they are all looking the same way. The results were good, but they were very similar to Ronan’s original skulls. We have the show chopped up into different epochs, and that is Epoch One, training directly on his skulls.
For Epoch Two I thought about how the coolest part about using GANs is that your getting a weird machine viewpoint of artwork. But feeding in all the skulls with the same layout is sort of like you are telling the machine how to look at the paintings. You’re giving it a very fixed perspective and a very normal perspective that we have already seen before.
So for Epoch Two, I basically played around with feeding the machine the skulls completely independent of any rotation or perspective, so the machine sees skulls that are all flipped around and stretched out. I’m using the same model, but the number of skulls in the training set jumped from 500 to 17,000 skulls. And the results are really, really good. It makes these really strange images that you would never expect. You can tell that they are skulls, but they really are not familiar. Ronan really loves those. He really likes to correct some of the skulls. He’ll say something like, ‘I like this one but it’s not right,’ or ‘There is never an image I am completely satisfied with,’ so he corrects it. He also does interpretations of them.
I also think that the Epoch Two skulls raise very interesting questions about authorship - since the network has learned exclusively from Ronan, but the outputs don't strongly resemble his work.
I asked Robbie about Ronan’s initial reaction to his work and how the relationship played out.
We are like opposites. He does not like the fact that my work is digital. He said the pixel is sad. And he really was skeptical about it. And right after I visited Paris, he was a little bit hesitant if he wanted to do the show because the French painters have the conception of technology and capitalism being the enemy. But now he is really excited about the show. But I think what is important to remember is that this is more like a confrontation than a collaboration. There are collaborative parts of it, but we really are sort of at odds.
Ronan explained to me that at first, he could not see where Robbie was making any decisions in the AI process. Like many, he thought that the “AI” and the “machine” was doing all the work and making all the choices. But quickly after working with Robbie and seeing that there is “choice and desire” in his work, he decided “the pixel is no longer sad.” But adds Ronan:
Of course it is not the same, I am not expecting the same thing from AI as I am from a painting. Both worlds are contiguous, but not the same. They are not the same rules. I hate the very idea of naturalism. As if everything was equivalent to everything else. I love the idea that there are two sets of rules, which allow us to play differently.
Ronan also pointed out that he does not keep all of his skull paintings. He curates them and many times he paints over the ones he does not like. He sees this curation process as not entirely unlike Robbie’s process of choosing which of the AI skulls to keep from the nearly unlimited number he can produce using GANs.
While the two have come to understand and respect each other’s working methods, there is a lot of interesting dialogue between them on what is an actual painting vs. something that is just an image of a painting. According to Ronan:
There is always difference between a painting and an image of a painting. And now [using GANs] there is an image of a painting that does not exist.
Sometimes I dream about the painting I want to do, and when I have done it, it is completely different. This indicates the direction, but you have to make your own way. And that is why the paintings will be presented as one by Ronan and then one by Robbie. Because then they become a mirror. And the question is, who is mirroring who? Originally they were skulls, but they become real vanities because of this idea of the mirror. With traditional vanities there is always a skull in the mirror which gives you the idea of time passing. Originally when I showed my skulls, each one was a painting on its own. But when paired with the works by Robbie, it creates a kind of double.
Simon Renard de St. André, Vanitas. Unknown.
Interestingly, Robbie agrees with Ronan that the individual images being produced by the GAN are just images of paintings (and in some cases, images of paintings that do not exist). But Robbie adds that he sees the trained GAN itself as the artwork. According to Robbie:
Ronan is right when he says that the AI skulls are "images of artwork" instead of artworks themselves. In my opinion, the actual artwork is the trained GAN itself, and the outputs are really just fragments or little glimpses of that (the trained GAN is almost just a compressed version of all the possible AI skulls).
Robbie often compares his process of working with GANs to that of the artist Sol LeWitt who is famous for writing out “rule cards” or algorithms for humans to execute to create his drawings.
Sol LeWitt Rule Card
Robbie explains:
The Sol LeWitt metaphor applies in multiple ways in GAN art. The data set is like the rule card, with rules created through curation - and the network interprets these to make art. But additionally, the network itself is also like the rule card, and the individual generations are just different interpretations/executions of those rules. This is in line with the idea that the individual works are just "tokens" of something larger - they're shadows of the network, the actual artwork.
At the same time, if the network itself is the piece of art, it's a very strange one, since it cannot be viewed or comprehended entirely (unlike the set of rules responsible for traditional generative artworks). We can only get small glimpses of it at a time. I'm not aware of any other type of art where this is true.
I have a lot of admiration for Ronan and his work - it seems almost unfair to Ronan to compare his work to the "images of artwork" output by the network. There's something present in the process of a traditional painter that I feel I'm missing as an artist - I'm not sure if it's dedication, rigor, the use of simple tools and not some complex machine, or something else entirely. Without being overly dramatic, there is something very honorable about how a very traditional painter operates; especially today when everything else is surrounded by technology. In short, I think that if I had to choose between the two types of skulls regardless of process or context, I would choose Ronan's skulls as my favorite. At the same time the Epoch Two AI skulls raise so many questions that I'm interested in - so including process/context, I'm more interested in them.
I’m an artist, I make work. But I am not the best at art history, I don’t have any traditional training, I don’t know how to paint or sketch or anything like that. I definitely do sympathize with Ronan’s view of digital work. Maybe he has seen a lot of low-quality digital work or he just doesn’t like the medium. It makes me wish that I was better at non-digital art.
I asked Ronan if he sees Robbie’s work as art or as inspiration for art.
Robbie introduced his own decisions and desires and changed the training images and the algorithms to make the work closer or further from the work I have done so far. It’s always interesting to bring something from outside the box into the realm of art. In the beginning, that can be seen as a threat. But in the end, it helps whatever is going on. If there is choice, if you can dream a little, it’s art. The skulls lend themselves well to AI and art because of the idea of the vanity of death. They therefore remain in ambiguity. And it is a disturbing ambiguity, the uncanny. Some will say it is about death and some will say it is about whatever, but I like maintaining this ambiguity in art. In the beginning I was worried that it was not possible to be free with AI. You can never say “that is not art, it is only a tool.” You have to find how to be free every time.
I asked Robbie how he finds this “freedom” in GANs and what makes good GAN art. He shared:
I really don’t like work that relies too heavily on the medium, like a watercolor painting where the whole interesting thing about the painting is that it is a watercolor and it relies on watercolor effects. My mom always called those “medium turds” or “watercolor turds.” I think the same applies to GANs where if it is reliant on the medium and the medium is the cool thing, then that’s not really art - it’s more like a tech demo. I think that the people that are making really cool work with GANs are using it in ways that are not obvious.
For example, in the show we have a box with a peephole in it, and when you look in, it will generate a skull and it will display it for like five seconds and then it will add an input vector to the “do not use list.” So basically you are going to be the only person to ever see that skull… ever. I think that is cool because it’s different and it’s new and it’s not too reliant on the GAN just being a GAN.
You Can’t Hand Someone an Apple and Call Yourself a Chef.
Not only is the artwork from Infinite Skulls of higher quality than anything I have seen from AI so far, the confrontation between the two artists and the resulting work forged through their conflict are the perfect visual symbol for the clash between AI and the traditional world at large.
I rarely anthropomorphize artificial intelligence and machine learning and prefer to think of these new technologies as augmenting human capabilities rather than replacing humans. But others have pushed me, asking, “Who is augmenting who?” in the relationship between AI and artists. If the relationships between AI and humans is symbiotic, then who is the host and who is the parasite? Though it may sound harsh, I think it is natural that people should ask themselves a similar question of the relationship between Ronan and Robbie, even if there is no clear answer.
While the two artists end up getting along and respecting each other’s methods in the end, each has to see the other as fuel or a raw material or ingredient to consume for their own artistic self-preservation. In both cases the artists are actively consuming the others work into their own as an ingredient, which is a different relationship than mere inspiration.
Ronan frames Robbie’s work as “photos of paintings that do not exist yet”, ostensibly because he himself has yet to create them, emphasizing that he is not happy with any of the works Robbie’s GAN produces until he “corrects” them. Note that Ronan also called Robbie and his AI “a guest in the studio” several times during our interview, which suggests a more passive role than an that of an equal in artistic collaboration. To further explain this relationship, Ronan explains, “It was like having a new guest in a jazz club,” again casting Robbie as a guest, or a “muse”, and not as a member of the band on the stage.
Similarly, Robbie has to treat Ronan’s 500 hundred skull paintings like unrefined wheat, grinding them down and further refining them to sufficiently anonymize them. He writes a program to randomize Ronan’s painting by stretching and flipping them to generate a less recognizable set of 17K training images from the initial 500 works before he can create art that is sufficiently different from Ronan’s to call it his own. Both must make a sacrifice of the other to produce their own work.
Ronan is rightfully proud to have painted two thousand skulls in the last two decades, but Robbie and his GAN can produce billions of skulls seemingly overnight, transforming Ronan into a sympathetic, man vs. machine, John Henry-like character.
It’s tempting to cast the story as two artists who overcome their many differences (age, language, tools) and some initial friction to collaborate on works that are as much by one as by the other. But to ignore the dynamic tension between the two artists is to miss much of what is interesting in the work. It is fitting that they landed on the theme of the skulls as vanities (traditional artworks designed to remind us of our own mortality) as it serves as an excellent thematic umbrella. After all, we all eventually return to the soil, only to become the ingredients in someone else’s narrative.